• Global Sites

Read noise in CMOS cameras: only rms is meaningful


Biologists who want to understand why camera makers throw around terms like median and rms.


An explanation, with visuals.

Learn more...

Learn more about noise and CMOS technology, read Changing the Game.

With any statistical parameter there are multiple models available to apply to the data. The classic electrical engineering method for calculating read noise is to define the root mean square (rms). This has always been the method used to calculate read noise for CCDs. Median and rms are both perfectly valid statistical models, but only rms noise accurately represents the experience that a user can expect from a camera. With CCDs there are never any issues regarding which model to use because the typical read noise for all pixels is very similar, thus rms and median are equivalent. With sCMOS, the structure of the sensor inherently has more pixel variation, and the extreme low noise of the sensor makes variation more statistically significant. So when it comes to evaluating camera performance, the truly meaningful spec is rms noise. The rms noise value provides insight into image quality as well as being the appropriate noise variable in quantitative calculations. For example, SNR measurements made empirically align with theory only when these simulations are done using rms noise values. Currently there is no industry standard in life science imaging for reporting noise specifications and it has become common practice for sCMOS to be specified by median read noise values. We include median noise data to facilitate superficial comparison with other sCMOS cameras, but we encourage users to be skeptical of median noise as a specification and to demand the more meaningful rms noise. The ORCA-Flash4.0 V2 Gen II sCMOS has 1.9 e- rms and 1.3 e- median typical read noise at standard scan.

All pixels or some pixels?
Readout Noise Distribution
The median value shown is simply the point at which half the pixels have more read noise and the other half have less. Given the nature of noise distribution in sCMOS cameras it is not particularly informative. RMS is the root mean square value of the read noise across all pixels and offers meaningful insight into image quality with pixel correction OFF. It is the value best used in image SNR calculations.

RMS or median noise values are valid only if all the pixels in the sensor are used or if the exclusion of outlier pixels is documented and explained. For the ORCA-Flash4.0 V2, we calculate both the rms and median read noise using every pixel in the sensor. This is done without any pixel correction functions or prequalification of the data. Since one goal of providing a spec is to enable accurate quantification of imaging results, this approach is consistent with our goal of providing the best quantitative scientific cameras.

Inquiry about Life Science Cameras

Contact us

Scroll to Top